Xref: alice comp.databases.oracle.server:67812 Path: alice!news-feed.fnsi.net!oleane!news-raspail.gip.net!news.gsl.net!gip.net!grolier!dispose.news.demon.net!demon!news.demon.co.uk!demon!jlcomp.demon.co.uk!not-for-mail From: "Jonathan Lewis" Newsgroups: comp.databases.oracle.server Subject: Re: log_buffer settings Date: Mon, 27 Sep 1999 08:58:30 +0100 Message-ID: <938419541.20161.0.nnrp-09.9e984b29@news.demon.co.uk> References: <7slfc7$bcb$1@nntp3.atl.mindspring.net> <5l3uNxd7fFLf02fzPpHtKmWtNBkT@4ax.com> <37eed2c6.31162369@news.gte.net> X-Trace: news.demon.co.uk 938419541 nnrp-09:20161 NO-IDENT jlcomp.demon.co.uk:158.152.75.41 X-Complaints-To: abuse@demon.net X-Newsreader: Microsoft Outlook Express 4.72.3110.5 X-MimeOLE: Produced By Microsoft MimeOLE V4.72.3110.3 Lines: 66 >Can you guess why so many "noted" authors and tuning experts mention >this in their books? How did this myth get started? It originally appeared in the manuals. In fact I think I will risk contradicting Thomas Kyte (just a little bit) on this one. I believe the 'redo log space request' figure is actually reporting the number of times that space was not made available in the log buffer - however (in full agreement with TK) it is usually the case that the space cannot be made available because the buffer has filled very rapidly and has to be written out to disc first. In the days of Oracle 6, and slower hardware, I often saw example where increasing the size of the buffer would reduce the number of waits quite significantly - but still leave a handful of waits that matched the number of log switches pretty closely. >Over the years, I'm convinced that the top 3 Oracle tuning books are >only 85% correct. Optimist !! Tuning books suffer from 3 main types of error. Ignorance - the author gets it wrong (sometimes because they read the manuals) Omission - there's lots of stuff that doesn't get in because of lack of space/time Aging - Oracle keeps moving on. Most of the errors in most of the tuning books are due to the fact that the product has changed since the book was written, and a good suggestion for versions x.y.z is either very stupid, or highly misleading for version a.b.c Postscript: I've just noticed on an 8.0.5 instance running under NT 4.0 that Oracle reports the log buffer space as 64K from v$sgastat and 72K from Show SGA even though v$parameter reports the setting in init.ora as the 8K set by one of the Oracle suggested setups. -- Jonathan Lewis Yet another Oracle-related web site: http://www.jlcomp.demon.co.uk johnvue@gte.net wrote in message <37eed2c6.31162369@news.gte.net>... > > >On Sun, 26 Sep 1999 14:00:04 -0400, Thomas Kyte >wrote: > >> >>reduce that log buffer size, get it back down into KILOBYTES, not megabytes. >> >>increasing the log buffer will not help (and can in fact hurt). >> >>the redo log space requests are not waits for space in the redo log buffer - >>they are waits for space in the online redo logs themselves. (some documentation >>erroneously tells you to increase the log_buffer -- it is wrong, increasing the >>log_buffer -- as you have seen -- will have no effect on this). >